What's available

Everything here can be required by your plugin, added to your agent, or called from your flows. Require a plugin by its id to get its tools, skills and subagents; never copy them.

Built into CLIO

These plugins ship inside CLIO, in the clio-builtin marketplace, and are always trusted.

Plugin	What it brings
`core`	Carried by every agent and never removable. Requires `fleet` and `streaming`. The `react` and `workflow` flows; the operating guidelines, rendering and task-lifecycle prompt fragments; the `planning` skill; the `clio-workspace` interactive catalog; and the conversation tools listed below.
`memory`	Long-term memory of the person, their projects and how they work, and recall of past conversations. Carried by the base `Agent`, and removable.
`fleet`	The models work can run on and what each account has left. The `update-models` skill.
`orchestration`	Requires `delegation`. The `run_flow` and `run_workflow` tools that start flows and workflows as background work, `make_plugin`, which writes a flow plugin the person can review and trust, the `council` and `arena` flows, and the `author-a-flow` skill.
`streaming`	Required by `core`, so every agent has it. Streams: named tables whose newest rows a script writes many times a second and a dashboard shows live. The `open_stream`, `read_stream` and `stop_stream` tools, the `stream` script binding, the `live-status` skill and its prompt fragment. See Watch something live.

Flows

Flow	From	What it does	Options
`react`	`core`	The default: the agent reads the conversation, calls tools and decides its next step until it answers.
`workflow`	`core`	Runs a workflow: steps that start as soon as the steps they wait on finish, with optional editors.	None; its input holds `workflow`, `task`, `editors`, `max_added_steps`, `max_edit_rounds`, `allowed_targets` and `approve_edits`
`council`	`orchestration`	Several members answer the same question independently, then one synthesis says where they agree and resolves where they differ.	`members` (targets; default three of `self`), `perspectives` (one per member), `models` (one per member), `synthesizer` (a target; empty synthesizes inside the flow), `synthesizer_model`
`arena`	`orchestration`	Contestants attempt the same task, then a judge compares the entries and picks one, with its reasons. The votes are appended to `arena/votes.jsonl` in the workspace.	`contestants` (targets; default three of `self`), `models` (one per contestant)
`deep-research`	`deep-research` in `clio`	Plans research directions, researches them in parallel, has an independent critic verify every claim against fetched sources, researches the gaps it names, then writes a report separating verified facts, supplied context and proposals.	`max_directions` (default 6, at most 12), `max_rounds` (gap rounds, default 2, at most 5), `concurrency` (researchers at once; 0 runs a round together), `researcher_model`, `critic_model`

Call any of them from a flow with ctx.call("council", question) or ctx.call(Flow("arena", models=[...]), task), start them from the model with run_flow, or use them as a workflow step.

Plugins in the clio marketplace

The clio marketplace is installed by default and trusted at install.

General capabilities

Plugin	What it brings	Requires
`workspace`	Read, search and edit files in the conversation’s workspace, and view images.
`shell`	A shell in the workspace, with CLIO’s managed execution runtime.
`ssh`	Run commands on a configured remote host over SSH.	`shell`
`slurm`	Submit Slurm batch jobs, record their ids, wait in bounded steps, and read a failed job’s output before resubmitting (skill).	`shell`
`web`	Fetch web pages and search the web (tool server).
`browser`	The shared browser: open, read and act on web pages from a script, shown to you in the side panel. See Use the browser.
`marimo`	Start a marimo notebook, open it in the shared browser, and work in its live Python session. Not in any agent by default.	`shell`
`delegation`	Start general subagents: any installed agent, or one of this agent’s defined subagents, with a task prompt.
`visuals`	Interactive analysis, dashboards, diagrams, 3D geometry, and rendered visual review in the side panel.
`connected-data`	Connect research data sources and work with the National Data Platform and knowledge stores.
`clio-search`	Search and read the connected CLIO service’s conversations, passages and workspace files.
`weather`	Weather forecasts for a place.

Documents

Plugin	What it brings	Requires
`documents`	A pack of `pdf`, `docx`, `xlsx`, `pptx`, `latex` and `opendocument`.
`document-runtime`	CLIO’s managed document toolchain: converters, fonts and bounded inspection.	`shell`
`pdf`	Read, review and produce PDF documents.	`document-runtime`
`docx`	Create, edit and review Word documents.	`document-runtime`
`xlsx`	Create, edit and analyze spreadsheets and CSV or TSV tables.	`document-runtime`
`pptx`	Create, edit and review presentations.	`document-runtime`
`latex`	Author LaTeX documents and build them to PDF.	`document-runtime`
`opendocument`	Create and edit OpenDocument text, spreadsheets and presentations.	`document-runtime`
`document-production`	Edit the canonical source, preserve packages, validate the format, review the rendered result, and apply review comments.	`documents`, `document-runtime`

Data formats and sources

Plugin	What it brings	Requires
`hdf5`	Open and inspect HDF5 files (tool server), with the `hdf5` skill.
`adios`	Inspect ADIOS2 BP4 and BP5 output (tool server), with the `adios2` skill.
`parquet`	Summaries, slices, previews and aggregates of Parquet files (tool server), with the `parquet` skill.
`pandas`	Load, profile, filter and summarize tables (tool server).
`plot`	Static plots: line, scatter, histogram, heatmap and time series (tool server).
`ndp`	Search National Data Platform datasets and stage their resources (tool server).
`geo`	Geocode places, filter points by radius and query geospatial features (tool server).
`data-onboarding`	Habits for unfamiliar datasets: inventory and audit tables, an experiment card, and an audit by a child agent (skills).

Science methods

Plugin	What it brings	Requires
`scientific-method`	Study design, literature review, research gaps, citations, data analysis, statistics, uncertainty, figures and scientific writing (skills), and the `data_analysis_expert` subagent.
`deep-research`	The `deep-research` flow with its `researcher` and `critic` subagents.	`delegation`
`geodesy`	GNSS position time series, station catalogs, region resolution and station-coverage comparison (skills).	`geo`
`phenotyping`	Plant-phenotyping methods: design and identity checks, growth traits, treatment response, spectral signals and reports (skills).	`data-onboarding`
`manufacturing`	A pack of `materials`, `manufacturing-processes`, `characterization`, `mechanical-testing`, `fatigue` and `fea-methods`.
`materials`	Materials science for metallic parts (skills), and the `materials_scientist` subagent.
`manufacturing-processes`	Additive and conventional processes, parameters, build orientation and manufacturability (skills), and the `manufacturing_expert` subagent.
`characterization`	X-ray CT, porosity, surface roughness, fractography and defect synthesis (skills), and the `characterization_expert` subagent.
`mechanical-testing`	Test design against standards, tensile testing, load frames and data acquisition (skills), and the `mechanical_testing_expert` subagent.
`fatigue`	Fatigue and failure: high-cycle, multiaxial, critical-plane and defect-driven life prediction, and fracture mechanics (skills), and the `fatigue_failure_expert` subagent.
`fea-methods`	Finite-element and topology-optimization method, independent of any program (skills).
`abaqus`	Script, run and audit Abaqus and Tosca packages, with result views (skills), and the `abaqus_engineer` subagent.	`shell`
`morphorm`	The Morphorm topology-optimization program, with bundled checks (skill).	`shell`

Agent-specific plugins

These carry one agent’s mission and are rarely useful elsewhere: factorio (with the evidence_researcher, research_methodologist, simulation_methodologist, virtual_lab, independent_reviewer, evidence_leaf and evidence_critic subagents), earthscope, appl-core, clio-data and spotter.

Agents in the clio marketplace

Agent	Name	Builds on	Adds
`clio/standard`	CLIO	`Agent`	`workspace`, `shell`, `web`, `documents`, `visuals`, `connected-data`, `delegation`, `orchestration`, `browser`
`clio/science`	CLIO Science	`standard`	`scientific-method`
`clio/factorio`	Factorio	`science`	`manufacturing`, `abaqus`, `morphorm`, `slurm`, `factorio`
`clio/earthscope`	EarthScope	`science`	`earthscope`, `geodesy`, `geo`, `ndp`
`clio/clio-data`	CLIO Data	`science`	`clio-data`, `data-onboarding`, `hdf5`, `adios`, `parquet`
`clio/appl-core`	OPAL	`science`	`appl-core`, `data-onboarding`, `phenotyping`, `parquet`
`clio/deep-researcher`	Deep Researcher	`Agent`	`orchestration`, `deep-research`
`clio/document-production`	Document Production	`Agent`	`document-production`, `workspace`, `shell`
`clio/clio-search`	CLIO Search	`Agent`	`clio-search`
`clio/spotter-ai`	SPOTTER	`Agent`	`spotter`

An agent in your own marketplace can carry any of these plugins by id. To start from one of these agents without writing a class, change a conversation’s plugins and save it as your own agent.

Tools by plugin

Each tool belongs to one plugin. Require that plugin to give your plugin or subagent the tool.

Plugin	Tools
`core`	`query_tasks`, `observe_tasks`, `wait_tasks`, `cancel_tasks`, `get_task_result`, `send_agent_message`, `pause_agent`, `resume_agent`, `restart_agent`, `ask_user`, `plan_enter`, `plan_update`, `plan_exit`, `write_todos`, `create_artifact`, `raise_alert_card`, `recall_context`, `load_skill`, `search_tools`, `run_code`, `set_goal`, `wait_for`, `goal_status`, `set_loop`, `loop_wakeup`, `cron_create`, `cron_list`, `cron_delete`, `workspace_resource_list`, `workspace_resource_inspect`, `workspace_resource_wait`, `workspace_resource_read`, `workspace_resource_search`, `workspace_resource_structure`
`memory`	`memory_search`, `memory_read`, `create_memory`, `update_memory`, `forget_memory`, `propose_memory_change`, `session_search`, `session_read_summary`, `session_read_context_frame`
`fleet`	`list_models`, `usage`, `refresh_provider_models`
`streaming`	`open_stream`, `read_stream`, `stop_stream`
`orchestration`	`run_flow`, `run_workflow`, `make_plugin`
`workspace`	`fs_read_file`, `fs_list_directory`, `fs_search_text`, `fs_write_file`, `fs_edit_file`, `view_image`
`shell`	`shell_bash`, `prepare_execution_runtime`
`ssh`	`ssh_execute`
`delegation`	`spawn_agent_task`, `spawn_agents_parallel`, `spawn_skill_task`
`visuals`	`create_a2ui_surface`, `update_a2ui_components`, `update_a2ui_data_model`, `delete_a2ui_surface`, `inspect_a2ui_surface`, `capture_a2ui_surface`, `publish_dashboard_report`
`connected-data`	`connected_data_status`, `connected_data_connect`, `connected_data_index`, `connected_data_open`, `connected_data_download`, `connected_data_write`, `connected_data_search_ndp`, `connected_data_ndp_organizations`, `connected_data_ndp_endpoint_connect`, `connected_data_ndp_endpoint_operations`, `connected_data_ndp_upload_artifact`, `connected_data_ndp_endpoint_request`, `knowledge_open`, `knowledge_search`, `knowledge_related`, `knowledge_create`, `knowledge_edit`
`document-runtime`	`prepare_document_runtime`, `prepare_document`
`pdf`	`view_pdf`
`clio-search`	`clio_search`, `clio_search_read`
`weather`	`get_weather_forecast`

Plugins with a tool server (web, hdf5, adios, parquet, pandas, plot, ndp, geo, spotter) bring that server’s tools, named after the server: web_fetch and web_search from web, for example.

Hook events

A plugin’s hooks can run on tool_call, tool_result, tool_batch, prompt_submit, stop, answer, session_start, session_end, subagent_start, subagent_stop, pre_compact, before_model and after_model. See Hooks for what each can do.

Build your first pluginRequire these plugins from your own.

Agent referenceBuild an agent on these plugins.